Modeling safety of highway work zones with random parameters and random effects models

Abstract

This paper presents an investigation of traffic safety in highway work zones using detailed data obtained from the results of a survey of project engineers and existing datasets. The observations were organized in monthly clusters that correspond to individual work zones; and a two-level random parameters negative binomial model that reflected the structure of the observations was estimated. The safety effects of various work zone design and traffic management features were identified, including lane shift, lane split, and detour, whose safety effects have not been evaluated in past research. This new insight into highway work zone safety was accomplished thanks to the better data acquired and the improved statistical model. A fixed parameters negative binomial model with random effects then was estimated to check its viability as an alternative to the random parameters model when the sample's large size makes estimation of the latter challenging. From a practical standpoint, the marginal effects on crash frequency computed from the model with random effects were quite similar to those computed from the random parameters model. This result indicates that, at least in some cases, convenient fixed parameters models may be a practical alternative to random parameters models. Utilization of an entire sample to estimate these conventional models may further compensate a less advanced model specification. The obtained negative binomial model with random effects has become useful for programming police enforcement in highway work zones in Indiana.